High-precision human body acquisition via multi-view binocular stereopsis

A Robust Multi‐View System for High‐Fidelity Human Body Shape Reconstruction

“…The proposed calibration pipeline using the encoded corners of calibration objects and human features improves the efficiency of calibration and the robustness of the system in terms of accuracy. For efficiency, we test the times of conducting calibration in our system and [RZY∗20] which applied checkerboard calibration. To calibrate all cameras, [RZY∗20] takes about 120 minutes while our method only about 15 minutes which improves by 8 times.…”

“…For efficiency, we test the times of conducting calibration in our system and [RZY∗20] which applied checkerboard calibration. To calibrate all cameras, [RZY∗20] takes about 120 minutes while our method only about 15 minutes which improves by 8 times. Instead of repeated calibration used in [RZY∗20] for each new model, the refinement strategy using human features further saves time and improves calibration efficiency.…”

“…In the last decade, some more advanced techniques have focused on calibration [LMS16,SF16], visibility estimation [SZFP16, XT19], low‐texture regions [RZY∗20, XLS∗20], and mesh refinement integrated with shading cues [WVT12, WWMT11, WLDW11], to make the resulting 3D models as accurate as possible. Recently, PatchMatch MVS methods [GLS15, SZFP16], which discard the front‐parallel assumption and adopt the core idea of the PatchMatch algorithm [BSFG09], have shown great power in solving dense matching problems with high accuracy and efficiency.…”

“…To this end, inspired by [BBB∗10, RZY∗20], a robust passive‐vision system (RHF‐Human) is proposed to acquire and reconstruct accurate human body shapes robustly. The flowchart of the system is shown in Figure 1.…”

“…Human body shape acquisition and reconstruction are active research topics in many fields ranging from animation to fashion design, health care, and digitized virtual humans. There are two types of human body shape acquisition techniques: (1) dynamic acquisitions techniques [PRMB15,CCS∗15,BBLR15,LFB17] that are focused on the natural movements of humans, such as motion capture and surface deformation, and (2) static acquisitions techniques [BBB∗10, RZY∗20] used for accurate human body shape reconstruction. In recent decades, many static active‐vision systems [TZL∗12, LCK∗21] and model‐based techniques [GWBB09, KBJM18] have been proposed for accurate human body shape reconstruction.…”